A Method for Multisource Heterogeneous Data Fusion and Modeling in New Power Systems

A method for multi-source heterogeneous data fusion and modeling in new power systems is proposed to address the difficulty of most methods in dealing with the complex structure of multi-source data in new power systems. Firstly, time registration is performed on the power data, and improved local global projection((LGP)) is used to extract features and delineate the molecular space. Then, input the subspace data into the Hermite orthogonal basis feedforward neural network for modeling learning to output the fusion results of the sub dataset. Finally, based on the MapReduce framework, parallelize the data samples and use the Reduce mechanism to summarize the fusion results of the sub datasets to obtain the final result. Based on the experimental analysis of microgrid data connected to a wind and solar power plant, the results show that the relative error and mean absolute percentage error of the proposed method for node voltage and power fusion results are 0.008, 0.012, and 0.014, 0.019, respectively, and the execution time is shorter, making it suitable for future data modeling trends in power systems.